[Letux-kernel] [PATCH v6 3/4] power: supply: twl4030-charger: move irq allocation to just before irqs are enabled

[H. Nikolaus Schaller]

Hi Sebastian,

> Am 15.06.2017 um 13:47 schrieb Sebastian Reichel <sebastian.reichel at collabora.co.uk>:
> 
> Hi,
> 
> On Wed, Jun 14, 2017 at 11:25:55AM +0200, H. Nikolaus Schaller wrote:
>> This avoids a potential race if irqs are enabled and triggered too early
>> before the worker is properly set up.
>> 
>> Suggested-by: Grygorii Strashko <grygorii.strashko at ti.com>
>> Signed-off-by: H. Nikolaus Schaller <hns at goldelico.com>
>> ---
>> drivers/power/supply/twl4030_charger.c | 28 ++++++++++++++--------------
>> 1 file changed, 14 insertions(+), 14 deletions(-)
>> 
>> diff --git a/drivers/power/supply/twl4030_charger.c b/drivers/power/supply/twl4030_charger.c
>> index 1fbbe0cc216a..3bebeecb4a1f 100644
>> --- a/drivers/power/supply/twl4030_charger.c
>> +++ b/drivers/power/supply/twl4030_charger.c
>> @@ -984,6 +984,16 @@ static int twl4030_bci_probe(struct platform_device *pdev)
>> 
>> 	platform_set_drvdata(pdev, bci);
>> 
>> +	if (bci->dev->of_node) {
>> +		struct device_node *phynode;
>> +
>> +		phynode = of_find_compatible_node(bci->dev->of_node->parent,
>> +						  NULL, "ti,twl4030-usb");
>> +		if (phynode)
>> +			bci->transceiver = devm_usb_get_phy_by_node(
>> +				bci->dev, phynode, &bci->usb_nb);
>> +	}
>> +
> 
> The notifier is not that much different from an irq. I see
> the call can access at least the iio channel (so iio channel
> should be registered first).

Hm. I only see that it can schedule the worker (which means the worker should be
initialized first) but does not access iio by itself.

Can the worker run before probe() succeeds?

> I suspect worker and power-supply
> should also be initialized/registered first. Best location seems
> to be directly before requesting the irqs.

I think they should be initialized even before devm_usb_get_phy_by_node().

> 
>> 	bci->channel_vac = devm_iio_channel_get(&pdev->dev, "vac");
>> 	if (IS_ERR(bci->channel_vac)) {
>> 		bci->channel_vac = NULL;
>> @@ -1006,6 +1016,10 @@ static int twl4030_bci_probe(struct platform_device *pdev)
>> 		return ret;
>> 	}
>> 
>> +	INIT_WORK(&bci->work, twl4030_bci_usb_work);
>> +	INIT_DELAYED_WORK(&bci->current_worker, twl4030_current_worker);
>> +
>> +	bci->usb_nb.notifier_call = twl4030_bci_usb_ncb;
> 
> You should configure the notifier block *before* registering it.

Isn't the notifier block already configured by devm_usb_get_phy_by_node() - except for the pointer to the notifier_call?

So the only harmful thing I can see is that a notifier arrives with usb_nb.notifier_call still being uninitialized as NULL.

Is this harmful or not? I don't know.

I can only guess that it could equally well be set right after devm_usb_get_phy_by_node()
and I have reworked it that way.

It still works on GTA04 and OpenPandora with neither observing a regression nor an improvement.

So I will send a v7 with what I think should be most robust.

BR and thanks,
Nikolaus

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.goldelico.com/pipermail/letux-kernel/attachments/20170619/693229f0/attachment.html>